Toward Understanding the Impact of Artificial Intelligence on Labor PDF

PERSPECTIVE PER...

PERSPECTIVE PERSPECTIVE Toward understanding the impact of artificial intelligence on labor Morgan R. Franka, David Autorb, James E. Bessenc, Erik Brynjolfssond,e, Manuel Cebriana, David J. Demingf,g, Maryann Feldmanh, Matthew Groha, José Loboi, Esteban Moroa,j, Dashun Wangk,l, Hyejin Younk,l, and Iyad Rahwana,m,n,1 Edited by Jose A. Scheinkman, Columbia University, New York, NY, and approved February 28, 2019 (received for review January 18, 2019) Rapid advances in artificial intelligence (AI) and automation technologies have the potential to significantly disrupt labor markets. While AI and automation can augment the productivity of some workers, they can replace the work done by others and will likely transform almost all occupations at least to some degree. Rising automation is happening in a period of growing economic inequality, raising fears of mass technological unemployment and a renewed call for policy efforts to address the consequences of technological change. In this paper we discuss the barriers that inhibit scientists from measuring the effects of AI and automation on the future of work. These barriers include the lack of high-quality data about the nature of work (e.g., the dynamic requirements of occupations), lack of empirically informed models of key microlevel processes (e.g., skill substitution and human–machine complementarity), and insufficient understanding of how cognitive technologies interact with broader economic dynamics and institutional mechanisms (e.g., urban migration and international trade policy). Overcoming these barriers requires improvements in the longitudinal and spatial resolution of data, as well as refinements to data on workplace skills. These improvements will enable multidisciplinary research to Downloaded from https://www.pnas.org by 190.129.166.212 on December 13, 2024 from IP address 190.129.166.212. quantitatively monitor and predict the complex evolution of work in tandem with technological progress. Finally, given the fundamental uncertainty in predicting technological change, we recommend developing a decision framework that focuses on resilience to unexpected scenarios in addition to general equilibrium behavior. | | automation employment economic resilience future of work| Artificial Intelligence (AI) is a rapidly advancing form of While technology generally increases productivity, technology with the potential to drastically reshape US AI may diminish some of today’s valuable employ- employment (1, 2). Unlike previous technologies, exam- ment opportunities. Consequently, researchers and ples of AI have applications in a variety of highly educated, policy makers worry about the future of work in both well-paid, and predominantly urban industries (3), includ- advanced and developing economies worldwide. As ing medicine (4, 5), finance (6), and information technol- an example, China is making AI-driven technology the ogy (7). With AI’s potential to change the nature of work, centerpiece of its economic development plan (8). how can policy makers facilitate the next generation of Automation concerns are not new to AI, and examples employment opportunities? Studying this question is date back even to the advent of written language. In made difficult by the complexity of economic systems ancient Greece (ca. 370 BC), Plato’s Phaedrus (9) de- and AI’s differential impact on different types of labor. scribed how writing would displace human memory a Media Laboratory, Massachusetts Institute of Technology, Cambridge, MA 02139; bDepartment of Economics, Massachusetts Institute of Technology, Cambridge, MA 02139; cTechnology & Policy Research Initiative, School of Law, Boston University, Boston, MA 02215; dSloan School of Management, Massachusetts Institute of Technology, Cambridge, MA 02139; eNational Bureau of Economic Research, Cambridge, MA 02138; f Harvard Kennedy School, Harvard University, Cambridge, MA 02138; gGraduate School of Education, Harvard University, Cambridge, MA 02138; h Department of Public Policy, The University of North Carolina at Chapel Hill, Chapel Hill, NC 27599; iSchool of Sustainability, Arizona State University, Tempe, AZ 85287; jGrupo Interdisciplinar de Sistemas Complejos, Departmento de Matematicas, Escuela Politécnica Superior, Universidad Carlos III de Madrid, 28911 Madrid, Spain; kKellogg School of Management, Northwestern University, Evanston, IL 60208; l Northwestern Institute on Complex Systems, Northwestern University, Evanston, IL 60208; mInstitute for Data, Systems, and Society, Massachusetts Institute of Technology, Cambridge, MA 02139; and nCenter for Humans and Machines, Max Planck Institute for Human Development, 14195 Berlin, Germany Author contributions: M.R.F., D.A., J.E.B., E.B., M.C., D.J.D., M.F., M.G., J.L., E.M., D.W., H.Y., and I.R. designed research; M.R.F. performed research; M.R.F. and M.G. analyzed data; and M.R.F., D.A., J.E.B., E.B., M.C., D.J.D., M.F., M.G., J.L., E.M., D.W., H.Y., and I.R. wrote the paper. The authors declare no conflict of interest. This article is a PNAS Direct Submission. This open access article is distributed under Creative Commons Attribution-NonCommercial-NoDerivatives License 4.0 (CC BY-NC-ND). 1 To whom correspondence should be addressed. Email: [email protected]. This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1900949116. Published online March 25, 2019. www.pnas.org/cgi/doi/10.1073/pnas.1900949116 PNAS | April 2, 2019 | vol. 116 | no. 14 | 6531–6539  and reading would substitute true knowledge with mere data. Critics have complained that prospective studies lack validation, More commonly, historians point to the Industrial Revolution but retrospective studies also find that robotics are diminishing and the riots of 19th-century Luddites (10) as examples where employment opportunities in US manufacturing (17, 28) [although technological advancement led to social unrest. Two examples not in Germany (29)]. from the recent past echo these concerns. Wassily Leontief, winner of the 1973 Nobel Prize in Economics, Optimist’s Perspective. Optimists suggest that technology may noted in 1952, “Labor will become less and less important... substitute for some types of labor but that efficiency gains from More workers will be replaced by machines. I do not see that technological augmentation outweigh transition costs (30–34), new industries can employ everybody who wants a job” (11). and, in many cases, technology increases employment for workers Similarly, US Attorney General Robert F. Kennedy commented who are in not direct competition with it (19, 35) [although recent in 1964, “Automation provides us with wondrous increases of follow-up work suggests these are temporary gains (28)]. Fur- production and information, but does it tell us what to do with thermore, the skill requirements of each job title are not static and the men the machines displace? Modern industry gives us the actually evolve over time to reflect evolving labor needs. For ex- capacity for unparalleled wealth—but where is our capacity to ample, workers may require more social skills because those skills make that wealth meaningful to the poor of every nation?” (12). remain difficult to automate (20). Even if technology depresses However, despite these long-lasting and oft-recurring con- employment for some types of labor, it can create new needs and cerns, society underwent profound transformations, the economy new opportunities through “creative destruction” (36–38). For continued to grow, technology continues to advance, and workers instance, the replacement of equestrian travel with automobiles continue to have jobs. Given this history of concern, what makes spurred demand for new roadside amenities, such as motels, gas human labor resilient to automation? Is AI a fundamentally new stations, and fast food (39). concern from technologies of the past? Answering these questions requires a detailed knowledge that Unifying Perspectives. On one hand, multiple dynamics ac- connects AI to today’s workplace skills. Each specific technology company technological change and create uncertainty about the alters the demand for specific types of labor, and thus the varying future of work. On the other hand, experts agree that occupations skill requirements of different job titles can obfuscate technol- are best understood as abstract bundles of skills (18, 40) and that ogy’s impact. In general, depending on the nature of the job, a technology directly impacts demand for specific skills instead of worker may be augmented by technology or in competition with it acting on whole occupations all at once (16, 19, 35, 41). Therefore, (13–15). For example, technological advancements in robotics can a detailed framework that connects specific skill types to career Downloaded from https://www.pnas.org by 190.129.166.212 on December 13, 2024 from IP address 190.129.166.212. diminish wages and employment opportunities for manufacturing mobility (18, 42) and to whole urban workforces (40) may help to workers (16, 17). However, technological change does not neces- unify competing perspectives (Fig. 1C). Existing studies have ar- sarily produce unemployment, and, in the case of AI, cognitive gued theoretically that different skill types underpin aggregate technology may actually augment workers (18, 19). For instance, labor trends, such as job polarization (16) and urban migration (43, machine learning appears to bolster the productivity of software 44), but robust empirical validation is made difficult by the spec- developers while also creating new investment and manufactur- ificity of modern skills data and their temporal sparsity. ing opportunities (e.g., autonomous vehicles). Complicating mat- ters further, the skill requirements of occupations do not remain Overcoming Barriers to Forecasting the Future of Work static, but instead change with changing technology (19, 20). In this section we identify barriers to our scientific modeling of In the remainder of this article, we describe how these technological change and the future of work. Along with each competing dynamics combined with insufficient data might allow barrier, we propose a potential solution that could enable im- contrasting perspectives to coexist. In particular, we argue that provement in forecasting labor trends. We provide a summary of the limitations into data about workplace tasks and skills restricts these barriers and solutions in Table 1. the viable approaches to the problem of technological change and the future of work. We offer suggestions to improve data Barrier: Sparse Skills Data. Forecasting automation from AI re- collection with the goal of enriching models for workplace skills, quires skills data that keep pace with rapidly advancing technol- employment, and the impact of AI. Finally, we suggest insights ogy [e.g., Moore’s Law (45), robots in manufacturing (17), and that improved data could provide in combination with a method- patent production (46–48)]. While skill types inform the theory of ological focus on resilience and forecasting. labor and technological change (1, 18, 21, 49), standard labor data focus on aggregate statistics, such as wage and employment Contrasting Perspectives numbers, and can lack resolution into the specifics that distinguish Doomsayer’s Perspective. Technology improves to make human different job titles and different types of work. For example, pre- labor more efficient, but large improvements may yield deleteri- vious studies have empirically observed a “hollowing” of the ous effects for employment. This obsoletion through labor sub- middle-skill jobs described by increasing employment share for stitution leads many to worry about “technological unemployment” low-skill and high-skill occupations at the expense of middle-skill and motivates efforts to forecast AI’s impact of jobs. One study occupations (16, 35) (reproduced in Fig. 1A). These studies use assessed recent developments in AI to conclude that 47% of current skills to explain labor trends but are limited empirically to measuring US employment is at high risk of computerization (23), while a con- annual wages instead of skill content directly. While wages may trasting study, using a different methodology, concluded that a less correlate with specific skills, wage alone fails to capture the defining alarming 9% of employment is at risk (24). Similar studies have features of an occupation, and models focused on only cognitive and looked at the impact of automation on employment in other physical labor fail to explain responses to technological change (21). countries and reached sobering conclusions: Automation will As another approach, data on educational requirements can affect 35% of employment in Finland (25), 59% of employment in add resolution to employment trends (50–52). For instance, jobs Germany (26), and 45 to 60% of employment across Europe (27). that require a bachelor’s degree may identify cognitive workers 6532 | www.pnas.org/cgi/doi/10.1073/pnas.1900949116 Frank et al.  A B C Fig. 1. Motivating and describing a framework to study technology’s impact on workplace skills. (A) Following ref. 21, we use American Community Survey national employment statistics to compare the change in employment share (y axis) of occupations according to their average annual wage (x axis) during two time periods. Employment share is increasing for low- and high-wage occupations at the expense of middle-wage occupations. (B) Following ref. 15, we use data from the Federal Reserve Bank of St. Louis to compare US productivity (real output per hour) and Downloaded from https://www.pnas.org by 190.129.166.212 on December 13, 2024 from IP address 190.129.166.212. workers’ income (real median personal income), which have traditionally grown in tandem. The efficiency gains of automating technologies are thought to contribute to this so-called great decoupling starting around the year 2000. (C) A framework for studying technological change, workplace skills, and the future of work as multilayered network. (Left) Cities and rural areas represent separate labor markets, but workers and goods can flow between them. (Middle) Each location can be represented as an employment distribution across occupations. Connections between occupations in a labor market represent viable job transitions. Job transitions are viable if workers of one job can meet the skill requirements of another job [i.e., “skill matching” (22)]. (Right) Workers’ varying skill sets represent bundles of workplace skills that tend to be valuable together. Skill pairs that tend to cooccur may identify paths to career mobility. Technology alters demand for specific workplace skills, thus altering the connections between skill pairs. As an example, machine vision software may impact the demand for human labor for some visual task. These alterations can accumulate and diffuse throughout the entire system as aggregate labor trends described in A and B. who are less susceptible to automation. Ideally, educational in- help explain the variability in current automation predictions stitutions train workers to possess valuable skills that lead to that enable contrasting perspectives. higher wages (53). However, looking at education and wages While publicly available skills data are limited, the US De- alone has proven insufficient to explain stagnating returns on partment of Labor’s O*NET database has seen recent use in labor education (16, 54, 55) and slow wage growth despite increases in research (e.g., refs. 23, 41, and 64). O*NET offers many benefits national productivity (14, 15, 41) (Fig. 1B). including a detailed taxonomy of skills and more regular updates Improving data on the skills required to perform specific job than preceding datasets. In 2014, O*NET began to receive partial tasks may provide better insights than wages and education updates twice annually, which is a considerable improvement on alone. For example, previous studies have considered occupations the Dictionary of Occupational Titles, which was published in four as routine or nonroutine and cognitive or physical (21, 56–63) or editions in 1939, 1949, 1965, and 1977, with a revision in 1991. looked at specific types of skills in relation to augmentation and However, employment trends and changing demand for specific substitution from technology (18, 41). Increasing a labor model’s tasks and skills might change faster than O*NET’s temporal res- specificity into workplace tasks and skills might further resolve olution and skill categorization can capture. Complicating matters labor trends and improve predictions of automation from AI. As an further, advances in AI and machine learning may be changing the example, consider that civil engineers and medical doctors are nature of automation, thereby altering the types of tasks that are both high-wage, cognitive, nonroutine occupations requiring affected by technology (3, 65). many years of higher education and additional professional cer- Furthermore, studies often use O*NET data to construct ag- tification. However, these occupations require distinct workplace gregations of skills, such as information input or mental processes skills that are largely nontransferable, and these occupations are (40), rather than focusing on skills at their most granular level. likely to interact with different technologies. Wages and education— Methodological choices aside, O*NET’s relatively static skill tax- and even aggregations of workplace skills—may be too coarse onomy poses its own problems as well. For instance, according to to distinguish occupations and, thus, may obfuscate the differ- O*NET, the skill “installation” is equally important to both com- ential impact of various technologies and complicate predictions puter programmers and to plumbers, but, undoubtedly, workers of changing skill requirements. In turn, these shortcomings may in these occupations are performing very dissimilar tasks when Frank et al. PNAS | April 2, 2019 | vol. 116 | no. 14 | 6533  Table 1. Tabulating the current barriers to forecasting the future of work along with proposed solutions Barrier Potential solution Sparse skills data  Adaptive skill taxonomies  Connect susceptible skills to new technology  Improve temporal resolution of data collection  Use data from career web platforms Limited modeling of resilience  Explore out-of-equilibrium dynamics  Identify workplace skill interdependencies  Connect skill relationships to worker mobility  Relate worker mobility to economic resilience in cities  Explore models of resilience from other academic domains Places in isolation  Labor dependencies between places (e.g., cities)  Identify skill sets of local economies  Identify heterogeneous impact of technology across places  Use intercity connections to study national economic resilience they are installing things on the job (see Fig. 2A and SI Appendix, Granular skills data will help elucidate the micro-scale impact section 1 for calculation). More generally, any static taxonomy for of AI and other technologies in labor systems. For instance, the Downloaded from https://www.pnas.org by 190.129.166.212 on December 13, 2024 from IP address 190.129.166.212. workplace skills is not ideal for a changing economy: Should specifications of recent patents might suggest automatable types mathematics and programming be two separate workplace skills of labor in the near future (46–48), thus elucidating the impact of given that they are both computational? Conversely, is “pro- technological change at the granularity of workplace-specific gramming” too broad given the variety of existing software and tasks and skills. The distribution of skill categories within occu- programming languages? Perhaps it is more appropriate to specify pations and over individuals’ careers can reveal how occupational programming tasks or specific programming languages (see Fig. 2B skill requirements evolve. As an example, consider that occupa- for an example), especially given the rapid development of AI tions such as software developer dynamically change the skill and machine learning. Likely, the correct abstraction is situation- requirements in job listings (e.g., “programming” in the 1990s vs. dependent, but O*NET data offer limited flexibility. “Python,” “Java,” “Kubernetes,” etc. today) to reflect the tools A B Fig. 2. Since the skill requirements of occupations may inform opportunities for career mobility, abstract skill data may obfuscate important labor trends. (A) We use O*NET data to identify the characteristic skill requirements for truck drivers, plumbers, and software developers (see SI Appendix, section 1 for calculation). Individual skills may be unique to an occupation (e.g., operating vehicles) or shared between occupations (e.g., low-light vision). The skill of installation is required by both plumbers and software developers, but this skill may not mean the same thing to workers in these two occupations. Programming is a skill required by software developers, but the coarseness of this skill definition may hide important dynamics brought on by new technology, including AI. (B) For example, we provide the percentage of Google searches for coding tutorials by programming language. Trends are smoothed using locally weighted scatter plot smoothing (see SI Appendix, section 2 for calculation). The Python programming language is widespread in the field of machine learning. Therefore, the increased ubiquity of AI and, in particular, machine learning may contribute to Python’s steady growth in popularity. 6534 | www.pnas.org/cgi/doi/10.1073/pnas.1900949116 Frank et al.  and required specialization of the time. Understanding the dy- It is difficult to construct resilient labor markets because of the namics of specific skills combined with the incomes within oc- uncertainty around technology’s impact on labor. For instance, cupations can capture the marginal value of different skills despite designing viable worker retraining programs requires detailed the dynamic nature of work. knowledge of the local workforce, fluency with current technology, Online career platforms offer an example of the empirical and an understanding of the complex dependencies between re- possibilities facilitated by nontraditional and new data sources. gional labor markets around the world (70, 71). Technology typically These websites collect real-time data that reflect labor dynamics performs specific tasks and may alter demand for specific workplace in certain industries. Data from workers’ resumes can improve our skills as a result. These micro-scale changes to skill demand can understanding of education and careers, as well as identifying accumulate into systemic labor trends including occupational skill workers’ transitions between occupations and skill sets. Addi- redefinition, employment redistribution (e.g., job creation and tionally, job postings capture fluctuations in labor demands and technological unemployment), and geographic redistribution demonstrate changes in demand for specific skills. Combined, (e.g., worker migration). Forecasting these complex effects re- these two sources of skills data offer an adaptive granular view quires a detailed understanding of the pathways along which into the changing nature of work that may detail where labor these dynamics occur. disconnects exist. Access to these private data sources is currently As an emblematic example of these complex dynamics, consider restricted and typically requires a data-sharing agreement that the competition between human bank tellers and automated teller protects personally identifiable information and other proprietary machines (ATMs) (described in ref. 72). Unexpectedly, national em- information. Of course, personal privacy and issues of represen- ployment for bank tellers rose with the adoption of ATMs. One ex- tative sampling are inherent to these data, but increased access planation is demand elasticity: As ATMs decreased the operating could meaningfully augment currently available open data on cost of bank branches, more bank branches opened nationwide to employment and workplace skills. One potential solution is to meet rising consumer demand. Another more complicated reason is construct a secure environment for the sharing of detailed skills the accompanying shift in fundamental skill requirements from and career data that is similar to the recent Social Science One clerical ability to social and persuasive skills used by salespeople and partnership (69) (see https://socialscience.one). customer service representatives. The story of bank tellers and ATMs is only fully captured by connecting the job-level changes in occu- Barrier: Limited Modeling of Resilience. Recent studies show pational skill composition with the system-level dynamics of de- that historical technology-driven trends may not capture the mand brought on by increased efficiency. Accordingly, an updated AI-driven trends we face today. Consequently, some have con- framework for labor and AI must capture the interactions of micro- Downloaded from https://www.pnas.org by 190.129.166.212 on December 13, 2024 from IP address 190.129.166.212. cluded that AI is a fundamentally new technology (3, 65). If the scopic workplace skills in combination to produce macroscopic labor trends of the past are not predictive of the employment trends trends, such as employment shifts, job polarization, and workers’ from current or future technologies, then how can policy makers spatial mobility (for example, see Fig. 3B). maintain and create new employment opportunities in the face of Existing theory of the matching process between job seekers AI? What features of a labor market lead to generalized labor and job vacancies (22) provides a stylized description of the resilience to technological change? matching process that lacks resolution into skill demand. Mapping A B Fig. 3. Skill complementarity may define the structural resilience of a workforce and inform worker retraining programs. (A) As in climatology and ecology, the structural pathways constraining labor dynamics could determine the resilience of a labor market to changing labor skill demands. In this example, we connect occupation pairs with high skill similarity because skill similarity might indicate easier worker transitions between job titles. Borrowing from research on ecological systems (66), the density of connections between occupations could determine “tipping points” for aggregate employment in cities. (B) With recent concerns of automation (67, 68), which jobs might be suitable for paralegals and legal assistants if employment for these jobs diminishes? Better resolution into skill requirements could help identify occupations that rely on similar skills but also rely on skills that are removed from competition with technology. In this example, we identify characteristic skills using the O*NET database to find that paralegals rely on many shared workplace skills with human resource specialists. Human resource specialists rely on social skills, which are not easily automated (20). See SI Appendix, section 1 for skill calculations. Frank et al. PNAS | April 2, 2019 | vol. 116 | no. 14 | 6535  the space of skill interdependencies (e.g., Fig. 1C) could inform adaptable as well. Perhaps ironically, advanced AI techniques may training and job assistance programs by identifying which types of help. Tools from machine learning (ML) and natural language work—and which locations—may experience augmentation and/or processing (NLP) may capture the latent structure in complex high- substitution with new technology. The detailed skill requirements of dimensional data, thus making them ideal tools for the proposed occupations determine the career mobility of individual workers, and application [and other applications in econometrics (86)]. For ex- thus changes to the demand for certain skills have the potential to ample, NLP may be used to process historical skills data from the redefine viable career trajectories and worker flow between occu- Dictionary of Occupational Titles into a format akin to the modern pations (e.g., middle layer of Fig. 1D). Therefore, mapping the re- O*NET data. ML can be used on longitudinal job postings data to lationships between jobs and skills that produce employment identify trends in skill demands that may reflect changes in tech- opportunities is a vital step for policy makers in the face of nological ability. Combining these modern computational methods technological change. with relevant sources of data may foster new insights into labor In related domains, tools from network science have already dynamics at a high temporal resolution. In turn, these methodo- provided new insights into modeling (and minimizing) systemic logical improvements can bolster labor forecasts and policy risk (73) in global credit (74) and financial industries (75), fore- makers’ ability to respond to real-time labor trends. casting the future exports of national economies (76–78), map- ping worker flows between industries (79) and firms (80), and Barrier: Places in Isolation. The impact of AI and automation will charting the changing industrial composition of cities (81–83) and vary greatly across geography, which has implications for the la- municipalities (84). Therefore, identifying the pathways along bor force, urban–rural discrepancies, and changes in the income which labor dynamics (e.g., how skills determine workers’ career distribution (87). The study of AI and automation are largely fo- mobility) occur may provide similarly useful insights into the im- cused on national employment trends and national wealth dis- pact of AI on labor. Similar methods have been used to measure parity. However, recent work demonstrates that some places (e.g., ecological resilience based on the structure of mutualistic inter- cities) are more susceptible to technological change than others species interactions (66, 85). These methods often rely on the (17, 64). Occupations form a network of dependencies which size and density of interconnected entities to estimate systemic constrain how easily jobs can be replaced by technology (82, 88). resilience to species removal—perhaps analogous to diminishing Therefore, the health of the aggregate labor market may depend demand for a skill with new technology (e.g., Fig. 3A). on the impact of technology on specific urban and rural labor Mapping skill dependencies will require appropriate data- markets (73, 84). handling methods. The ideal skills data should reflect the dynamic Although technological change alters demand for specific Downloaded from https://www.pnas.org by 190.129.166.212 on December 13, 2024 from IP address 190.129.166.212. nature of skill representation, and so the methods we use to de- workplace tasks and skills, current skills data mask the specific skill tect, categorize, and measure the demand for skills must be sets that comprise and differentiate the workforces of different C D A B Fig. 4. A data pipeline that overcomes barriers to studying the future of work. (A) Inputs into the data pipeline include structured and unstructured data that detail regional variations in labor and granular skills data in relation to technological change. (B) Data from a variety of sources will need to be centralized and processed into a form that economists and data scientists can easily use (e.g., NLP to identify skill from resume and job postings). (C) Cleaned data feed a model for both the intercity (e.g., worker migration) and intracity (e.g., changes to local career mobility) labor trends brought on by technological change. (D) Outputs from this model will forecast the labor impact of technological change. These forecasts will inform policy makers seeking to implement prudent policy and individual workers attempting to navigate their careers. 6536 | www.pnas.org/cgi/doi/10.1073/pnas.1900949116 Frank et al.  geographies. In part, this is because skills data from nationwide Conclusion surveys, such as the O*NET database, average over the regional AI has the potential to reshape skill demands, career opportunities, variability in the required skills of workers with shared job titles. and the distribution of workers among industries and occupations in For example, software developers seeking employment in Silicon the United States and in other developed and developing countries. Valley may need to advertise more specific skill sets than similar However, researchers and policy makers are underequipped to employees in a shallower labor market (following the division of forecast the labor trends resulting from specific cognitive tech- labor theory). Exacerbating this trend, the same AI technologies that augment high-wage cognitive employment are more abun- nologies, such as AI. Typically, technology is designed to perform dant in large cities, while the physical low-wage tasks that are a specific task which alters demand for specific workplace skills. most readily replaced by robotics are more abundant in small The resulting alterations to skill demands diffuse throughout cities and rural communities. This observation suggests that na- the economy, influencing occupational skill requirements, career tional wealth disparity is reflected in the wealth disparity between mobility, and societal well-being (e.g., impacts to workers’ social large and small cities akin to wage inequality across individuals. identity). Identifying the specific pathways of these dynamics has Improved models for spatial interdependencies require more granular skills data (discussed above) and new insights into the been constrained by coarse historical data and limited tools for mechanisms that create today’s cross-sectional geographic trends. modeling resilience. We can overcome these obstacles, however, For instance, how do university towns, where people gain valuable by prioritizing data collection that is detailed, responsive to real- cognitive skills, contribute to the productivity of large cities? Do time changes in the labor market, and respects regional variability these economic connections help explain why university towns (see Fig. 4 for a data-pipeline schematic). Specifically, better ac- perform surprisingly well compared with similarly sized cities cess to unstructured skills data from resumes and job postings according to socioeconomic indicators [including exposure to along with new indicators for recent technological change (e.g., automation (64)]? Furthermore, just as internal connectivity determines urban patent data) and models for both intercity and intracity labor economic resilience (83), so too can the connections between US dependencies will enable new and promising techniques for un- cities underpin the economic health of the national economy (48). derstanding and forecasting the future of work. This improved For instance, an interruption in the supply chain of well-educated data collection will enable the use of new data-driven tools, in- cognitive workers may stifle an urban economy that normally at- cluding machine learning applications and systemic modeling tracts skilled workers. Therefore, it behooves policy makers to that more accurately reflects the complexity of labor systems. New Downloaded from https://www.pnas.org by 190.129.166.212 on December 13, 2024 from IP address 190.129.166.212. understand the connections between their local labor market data will lead to new research that enriches our understanding of and other urban labor markets to assess the resilience of their local economy. Since employment opportunities are central the impact of technology on modern labor markets. in people’s decision to relocate (43) and skill matching is es- sential to the job matching process (22), understanding the Acknowledgments This work summarizes insights from the workshop on Innovation, Cities, and the constituent skill sets in cities can inform models for the spatial Future of Work, which was funded by NSF Grant 1733545. This work was mobility of workers and improve our understanding of career mo- supported by the Massachusetts Institute of Technology (MIT) and the MIT bility and career incentives. Initiative on the Digital Economy. 1 Mitchell T, Brynjolfsson E (2017) Track how technology is transforming work. Nature 544:290–292. 2 Brynjolfsson E, Rock D, Syverson C (2017) Artificial intelligence and the modern productivity paradox: A clash of expectations and statistics. NBER Working Paper 22401 (National Bureau of Economic Research, Cambridge, MA). 3 Brynjolfsson E, Mitchell T (2017) What can machine learning do? Workforce implications. Science 358:1530–1534. 4 Barlow M (2016) AI and Medicine (O’Reilly Media, Sebastopol, CA). 5 Darcy AM, Louie AK, Roberts LW (2016) Machine learning and the profession of medicine. JAMA 315:551–552. 6 Dunis CL, Middleton PW, Karathanasopolous A, Theofilatos K (2016) Artificial Intelligence in Financial Markets: Cutting Edge Applications for Risk Management, Portfolio Optimization and Economics (Springer, New York). 7 Witten IH, Frank E, Hall MA, Pal CJ (2016) Data Mining: Practical Machine Learning Tools and Techniques (Morgan Kaufmann, Burlington, MA). 8 He ABY (2017) How China is preparing for an AI-powered future (Wilson Center, Washington, DC). 9 Yunis H (2011) Plato: Phaedrus (Cambridge Univ Press, Cambridge, UK). 10 Jones SE (2013) Against Technology: From the Luddites to Neo-Luddism (Routledge, Abingdon, UK). 11 Leontief W (1952) Machines and man. Sci Am 187:150–164. 12 Kennedy RF (1964) Welcoming remarks to the opening session of the Interfederal Assembly of Pax Romana (US Department of Justice, Washington, DC). Available at https://www.justice.gov/sites/default/files/ag/legacy/2011/01/20/07-20-1964.pdf. 13 Acemoglu D, Autor D (2011) Skills, tasks and technologies: Implications for employment and earnings. Handbook of Labor Economics (Elsevier, Amsterdam), Vol 4, pp 1043–1171. 14 Brynjolfsson E, McAfee A (2012) Race Against the Machine: How the Digital Revolution Is Accelerating Innovation, Driving Productivity, and Irreversibly Transforming Employment and the Economy (Digital Frontier, Lexington, MA). 15 Brynjolfsson E, McAfee A (2014) The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies (Norton, New York). 16 Autor D (2010) The polarization of job opportunities in the US labor market: Implications for employment and earnings (Center for American Progress and The Hamilton Project). 17 Acemoglu D, Restrepo P (2017) Robots and jobs: Evidence from US labor markets. NBER Working Paper 23285 (National Bureau of Economic Research, Cambridge, MA). 18 Autor DH, Levy F, Murnane RJ (2003) The skill content of recent technological change: An empirical exploration. Q J Econ 118:1279–1333. 19 Bessen JE (2016) How computer automation affects occupations: Technology, jobs, and skills. Law and Economics Research Paper 15-49 (Boston Univ School of Law, Boston). 20 Deming D, Kahn LB (2018) Skill requirements across firms and labor markets: Evidence from job postings for professionals. J Labor Econ 36:S337–S369. Frank et al. PNAS | April 2, 2019 | vol. 116 | no. 14 | 6537  21 Acemoglu D, Autor D (2011) Skills, tasks and technologies: Implications for employment and earnings. Handbook of Labor Economics (Elsevier, Amsterdam), Vol 4, pp 1043–1171. 22 Mortensen DT, Pissarides CA (1994) Job creation and job destruction in the theory of unemployment. Rev Econ Stud 61:397–415. 23 Frey CB, Osborne MA (2017) The future of employment: How susceptible are jobs to computerisation? Technol Forecast Soc Change 114:254–280. 24 Arntz M, Gregory T, Zierahn U (2016) The risk of automation for jobs in OECD countries: A comparative analysis. OECD Social, Employment and Migration Working Paper 189 (OECD Publishing, Paris). Available at www.ifuturo.org/sites/default/files/docs/automation.pdf. 25 Pajarinen M, Rouvinen P, Ekeland A (2015) Computerization threatens one-third of Finnish and Norwegian employment. ETLA Brief 34. Available at https://www. etla.fi/wp-content/uploads/ETLA-Muistio-Brief-34.pdf. 26 Brzeski C, Burk C (2015) Die Roboter kommen: Folgen der Automatisierung für den deutschen Arbeitsmarkt [The Robots are coming: Consequences of Automation for the German Labor Market]. INGDiBa Econ Res 30:1–7. German. 27 Bowles J (July 24, 2014) The computerisation of European jobs. Bruegel blog. Available at bruegel.org/2014/07/the-computerisation-of-european-jobs/. 28 Bessen J (2018) AI and Jobs: The role of demand. NBER Working Paper No. 24235. Available at https://www.nber.org/papers/w24235. Accessed March 15, 2019. 29 Dauth W, Findeisen S, Südekum J, Woessner N (2017) German robots – The impact of industrial robots on workers. IAB Discussion Paper 30/2017 (Institute for Employment Research, Nuremberg, Germany). 30 Naone E (November 11, 2009) The dark side of the technology utopia. MIT Technology Review. Available at https://www.technologyreview.com/s/416244/ the-dark-side-of-the-technology-utopia/. 31 Frase P (2016) Four Futures: Life after Capitalism (Verso Books, New York). 32 Farrell C (February 11, 2013) Will robots create economic utopia? Bloomberg. Available at https://www.bloomberg.com/news/articles/2013-02-11/will-robots- create-economic-utopia. 33 Avent R (September 19, 2016) A world without work is coming – It could be utopia or it could be hell. The Guardian. Available at https://www.theguardian.com/ commentisfree/2016/sep/19/world-without-work-utopia-hell-human-labour-obsolete. 34 Ip G (August 2, 2017) We survived spreadsheets, and we’ll survive AI. Wall Street Journal. Available at https://www.wsj.com/articles/wesurvived-spreadsheets- and-well-survive-ai-1501688765. 35 David H, Dorn D (2013) The growth of low-skill service jobs and the polarization of the US labor market. Am Econ Rev 103:1553–1597. 36 Aghion P, Howitt P (1990) A model of growth through creative destruction. NBER Working Paper 3223 (National Bureau of Economic Research, Cambridge, MA). 37 Aghion P, Howitt P (1994) Growth and unemployment. Rev Econ Stud 61:477–494. 38 Bartelsman E, Haltiwanger J, Scarpetta S (2004) Microeconomic evidence of creative destruction in industrial and developing countries (World Bank, Washington, DC). 39. Jackson K (April 25, 1993) The world’s first motel rests upon its memories. Seattle Times. Available at community.seattletimes.nwsource.com/archive/? date=19930425&slug=1697701. 40 Kok S, ter Weel B (2014) Cities, tasks, and skills. J Reg Sci 54:856–892. 41 MacCrory F, Westerman G, Alhammadi Y, Brynjolfsson E (2014) Racing with and against the machine: Changes in occupational skill composition in an era of rapid technological advance. Proceedings of the Thirty-Fifth International Conference on Information Systems. Available at https://dblp.org/rec/html/conf/icis/ MacCroryWAB14. Accessed March 15, 2019. Downloaded from https://www.pnas.org by 190.129.166.212 on December 13, 2024 from IP address 190.129.166.212. 42 Alabdulkareem A, Frank MR, Sun L, AlShebli B, Hidalgo C, Rahwan I (2018) Unpacking the polarization of workplace skills. Sci Adv 4:eaao6030. 43 International Organization for Migration (2015) World migration report 2015 - Migrants and cities: New partnerships to manage mobility (International Organization for Migration, Le Grand-Saconnex, Switzerland). 44 Fields GS (1975) Rural-urban migration, urban unemployment and underemployment, and job-search activity in LDCs. J Dev Econ 2:165–187. 45 Sneed A (May 19, 2015) Moore’s law keeps going, defying expectations. Scientific American. Available at https://www.scientificamerican.com/article/ moore-s-law-keeps-going-defying-expectations/. 46 Youn H, Strumsky D, Bettencourt LMA, Lobo J (2015) Invention as a combinatorial process: Evidence from US patents. J R Soc Interf 12:20150272. 47 Strumsky D, Lobo J (2015) Identifying the sources of technological novelty in the process of invention. Res Pol 44:1445–1461. 48 Strumsky D, Thill J-C (2013) Profiling US metropolitan regions by their social research networks and regional economic performance. J Reg Sci 53:813–833. 49 Weaver A, Osterman P (2017) Skill demands and mismatch in US manufacturing. ILR Rev 70:275–307. 50 Autor DH (2015) Why are there still so many jobs? The history and future of workplace automation. J Econ Perspect 29:3–30. 51 Goldin L, Katz C (2008) The Race Between Education and Technology (Harvard Univ Press, Cambridge, MA). 52 Autor DH, Katz LF, Kearney MS (2008) Trends in US wage inequality: Revising the revisionists. Rev Econ Stat 90:300–323. 53 Kerckhoff AC (2001) Education and social stratification processes in comparative perspective. Sociology of Education 74:3–18. 54 Cingano F (2014) Trends in income inequality and its impact on economic growth. OECD Social, Employment and Migration Working Paper 163 (OECD Publishing, Paris). Available at www.oecd.org/els/soc/trends-in-income-inequality-and-its-impact-on-economic-growth-SEM-WP163.pdf. 55 Beaudry P, Green DA, Sand BM (2016) The great reversal in the demand for skill and cognitive tasks. J Labor Econ 34:S199–S247. 56 Spitz-Oener A (2006) Technical change, job tasks, and rising educational demands: Looking outside the wage structure. J Labor Econ 24:235–270. 57 Antonczyk D, Fitzenberger B, Leuschner U (2009) Can a task-based approach explain the recent changes in the German wage structure? ZEW Discussion Papers 08-132 (ZEW – Leibniz Centre for European Economic Research, Mannheim, Germany). Available at hdl.handle.net/10419/27615. 58 Dustmann C, Ludsteck J, Schanberg U (2009) Revisiting the German wage structure. Q J Econ 124:843–881. 59 Firpo S, Fortin NM, Lemieux T (2011) Occupational tasks and changes in the wage structure. IZA Discussion Papers 5542 (Institute for the Study of Labor, Bonn, Germany). Available at ftp.iza.org/dp5542.pdf. 60 Ikenaga T (2009) Polarization of the japanese labor market: The adoption of it and changes in tasks contents. The Japanese Journal of Labour Studies 584:73–90. 61 Michaels G, Natraj A, Van Reenen J (2010) Has ICT polarized skill demand? Evidence from eleven countries over 25 years. NBER Working Paper 16138 (National Bureau of Economic Research, Cambridge, MA). 62